Designing New Protein Functions Using Deep Learning with David Baker

The Power of Protein Design

• 💡 Proteins are the fundamental workers in cells, mediating all critical life processes and solving evolutionary challenges.
• 🎯 Unlike traditional methods that modify existing proteins, the goal is to design entirely new proteins from scratch to address modern problems.
• 🧠 This involves navigating an astronomical number of possible amino acid sequences (10^130) to find functional designs, far beyond what exists in nature.

Deep Learning for De Novo Protein Creation

• 🚀 Early efforts used physically-based models to predict protein structures by estimating atomic interactions and finding lowest energy states.
• ⚡ The field transitioned to deep learning methods, specifically diffusion models, analogous to image generation AI like DALL-E.
• 🔬 These models are trained on 150,000+ known protein structures to progressively remove noise and generate novel, functional protein shapes.
• ✅ The key is conditioning the generative process to design proteins for specific desired functions or targets, such as binding to a particular structure.

Medical Innovations Through Protein Design

• 💊 Designed proteins can block snake venom toxins and activate the immune system for cancer treatment by bringing together cellular receptors.
• 💡 New proteins act as potent blockers for autoimmunity (e.g., TNF signaling) and can be designed to mimic antibody therapeutics.
• 🧠 Proteins are being developed to target and destroy misfolding proteins associated with neurodegenerative diseases like Alzheimer's (amyloid beta, tau protein).
• 💉 This technology led to the design of the first clinically approved COVID vaccine, utilizing self-assembling protein nanoparticles.

Advancing Technology and Sustainability

• 🛠️ De novo designed nanopores can be embedded in membranes or silicon nitride chips for advanced sensing and DNA sequencing.
• 🧩 Protein switches can change shape in response to external factors, enabling precise control over biological processes, like immune responses.
• 🌱 Proteins are being designed as catalysts for bond-making (e.g., tuberculosis drugs) and bond-breaking (e.g., plastic degradation).
• ✨ New proteins can template the growth of semiconductor materials (zinc oxide) and biominerals (hydroxyapatite) for novel material science.
• ☀️ Efforts include designing proteins to improve photosynthesis and enhance light harvesting for energy applications.

Future Directions and Challenges

• 📈 Current research focuses on developing atomic-level diffusion models for more sophisticated control over protein design, moving beyond amino acid residues.
• 📊 Building general biological models by training on diverse datasets, including metagenomes, is crucial for extending capabilities across various biological interactions.
• ⚠️ While powerful, the main challenge is ensuring designs work experimentally and addressing potential unintended consequences or immune responses in therapeutic applications.